> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usesynth.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Concepts

> Projects, runs, workers, context, artifacts, budgets, and approvals.

Managed Research is organized around durable work rather than chat turns.

## Project

A project is the reusable control unit for repo bindings, files, datasets, credentials, notes, knowledge, budgets, and policy. Use a project when workers need persistent context or repeated runs.

## Run

A run is one execution of a goal against the current project and launch configuration. Runs are inspectable through state, messages, logs, task counts, actor status, artifacts, checkpoints, usage, questions, approvals, and final reports.

## Worker

Workers execute research or engineering tasks inside managed workspaces. The orchestrator can create tasks, assign workers, coordinate reviewers, and synthesize outcomes depending on runbook and work mode.

## Evidence

Evidence is the durable record that makes Managed Research different from a chat transcript:

* runtime messages
* task and actor state
* logs and logical timeline
* checkpoints and branches
* artifact manifest and file outputs
* usage and budget state
* PRs, reports, and final outputs

## Launch configuration

Launch fields shape a run:

* `host_kind` chooses the execution substrate.
* `work_mode` chooses goal posture.
* `runbook` chooses collaboration posture.
* `providers` bind runtime provider capability.
* `agent_harness`, `agent_model`, and `agent_model_params` choose public agent runtime settings.

Backend preflight remains authoritative for whether a launch is allowed.

## Supported integration paths

Use MCP for agent-client workflows and the Python SDK for scripts. Direct `/smr` REST wiring is internal and unstable.